Weight of Evidence: Criteria and Applications.

Toxicologists and authorities evaluate substances that in the traditional way refer to data and knowledge on the toxic mechanism. Non-testing methods (NTMs) proved to be a valuable resource for risk assessment of chemical substances. Indeed, they can be particularly useful when the information provided by different sources is integrated to increase confidence in the result. In this chapter, we will address data obtained from NTM, including in silico models and read-across. Typically, these two approaches are used separately, but in this way, we lose some information. The integration of different results can be sometimes difficult because different methods can lead to conflicting results and because a clear guideline for integrating information from different sources was not available in the recent past. In this chapter, we present and discuss the recently published guideline from EFSA for integrating and weighting evidence for scientific assessment, and how to proceed, manually or through novel integrated tool, SWAN. Moreover, practical examples of the application of these integration principles on evidence from different in silico models are shown. These examples represent a demonstration of the suitability and effectiveness of in silico methods for risk assessment, as well as a practical guide to end users to perform similar analyses on likely hazardous chemicals.

Artificial Intelligence and Machine Learning Methods to Evaluate Cardiotoxicity following the Adverse Outcome Pathway Frameworks.

Cardiovascular disease is a leading global cause of mortality. The potential cardiotoxic effects of chemicals from different classes, such as environmental contaminants, pesticides, and drugs can significantly contribute to effects on health. The same chemical can induce cardiotoxicity in different ways, following various Adverse Outcome Pathways (AOPs). In addition, the potential synergistic effects between chemicals further complicate the issue. In silico methods have become essential for tackling the problem from different perspectives, reducing the need for traditional in vivo testing, and saving valuable resources in terms of time and money. Artificial intelligence (AI) and machine learning (ML) are among today's advanced approaches for evaluating chemical hazards. They can serve, for instance, as a first-tier component of Integrated Approaches to Testing and Assessment (IATA). This study employed ML and AI to assess interactions between chemicals and specific biological targets within the AOP networks for cardiotoxicity, starting with molecular initiating events (MIEs) and progressing through key events (KEs). We explored methods to encode chemical information in a suitable way for ML and AI. We started with commonly used approaches in Quantitative Structure-Activity Relationship (QSAR) methods, such as molecular descriptors and different types of fingerprint. We then increased the complexity of encoders, incorporating graph-based methods, auto-encoders, and character embeddings employed in neural language processing. We also developed a multimodal neural network architecture, capable of considering the complementary nature of different chemical representations simultaneously. The potential of this approach, compared to more conventional architectures designed to handle a single encoder, becomes apparent when the amount of data increases.

Virtual Extensive Read-Across: A New Open-Access Software for Chemical Read-Across and Its Application to the Carcinogenicity Assessment of Botanicals.

Read-across applies the principle of similarity to identify the most similar substances to represent a given target substance in data-poor situations. However, differences between the target and the source substances exist. The present study aims to screen and assess the effect of the key components in a molecule which may escape the evaluation for read-across based only on the most similar substance(s) using a new open-access software: Virtual Extensive Read-Across (VERA). VERA provides a means to assess similarity between chemicals using structural alerts specific to the property, pre-defined molecular groups and structural similarity. The software finds the most similar compounds with a certain feature, e.g., structural alerts and molecular groups, and provides clusters of similar substances while comparing these similar substances within different clusters. Carcinogenicity is a complex endpoint with several mechanisms, requiring resource intensive experimental bioassays and a large number of animals; as such, the use of read-across as part of new approach methodologies would support carcinogenicity assessment. To test the VERA software, carcinogenicity was selected as the endpoint of interest for a range of botanicals. VERA correctly labelled 70% of the botanicals, indicating the most similar substances and the main features associated with carcinogenicity.

Machine Learning for Efficient Prediction of Protein Redox Potential: The Flavoproteins Case.

Determining the redox potentials of protein cofactors and how they are influenced by their molecular neighborhoods is essential for basic research and many biotechnological applications, from biosensors and biocatalysis to bioremediation and bioelectronics. The laborious determination of redox potential with current experimental technologies pushes forward the need for computational approaches that can reliably predict it. Although current computational approaches based on quantum and molecular mechanics are accurate, their large computational costs hinder their usage. In this work, we explored the possibility of using more efficient QSPR models based on machine learning (ML) for the prediction of protein redox potential, as an alternative to classical approaches. As a proof of concept, we focused on flavoproteins, one of the most important families of enzymes directly involved in redox processes. To train and test different ML models, we retrieved a dataset of flavoproteins with a known midpoint redox potential (Em) and 3D structure. The features of interest, accounting for both short- and long-range effects of the protein matrix on the flavin cofactor, have been automatically extracted from each protein PDB file. Our best ML model (XGB) has a performance error below 1 kcal/mol (∼36 mV), comparing favorably to more sophisticated computational approaches. We also provided indications on the features that mostly affect the Em value, and when possible, we rationalized them on the basis of previous studies.